In science classrooms, students usually see and work with data that's intended to tell them right away about the natural world. Students then often treat the data we provide to them as factual, rather than as a source of evidence (Duschl 2008; Sandoval and Millwood 2005; Berland and Reiser 2009; McNeill and Berland 2017; Hancock, Kaput, and Goldsmith 1992; Manz 2016), and struggle to identify sources of error or uncertainty in evidence (Masnick and Klahr 2003). Rather than always aim for certainty, teachers need to let scientific work with data be more uncertain. Teachers can then allow students to do more of the "puzzling over data" themselves, whether it's deciding what sorts of data to collect and how, or developing criteria for what counts as "good data" (Ko and Krist 2018; Manz and Suárez 2018). To get students to do this sort of scientific thinking, and to prepare them for future work with messy data, teachers need to break data--occasionally, data must fail to tell students about the natural world. Sensor-based science labs are a great context for students to understand where data come from. To develop new types of sensor-based labs for high school biology, the authors designed many variations on traditional lab experiments. Their labs used "Do-it-Yourself" probeware (Tinker and Krajcik 2001), including low-cost commercial sensors, and internet-connected Raspberry Pi computers (Hsi, Hardy, and Farmer 2017). These sensors presented no more safety risk than other low-voltage electronics though the sensors themselves can be damaged (e.g., by water); they allow students to view the sensors as designed technologies rather than incomprehensible black boxes (Hardy, Dixon, and Hsi 2020). This article includes three effective ways to get students puzzling over their sensor data.